Research map / Graph learning / Graph algorithms and clustering
Graph-GRPO: Training Graph Flow Models with Reinforcement Learning
Baoheng Zhu, Deyu Bo, Delvin Ce Zhang, Xiao Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add RL training (Graph-GRPO) to the DeFoG base model (ablation) | DeFoG · PMO (23 tasks) | AUC-top10: 11.079 -> 16.901 | Improved |
| add refinement without RL (ablation) | DeFoG · PMO (23 tasks) | AUC-top10: 11.079 -> 15.251 | Improved |
| add refinement on top of RL (ablation) | Graph-GRPO · PMO (23 tasks) | AUC-top10: 16.901 -> 17.950 | Improved |
| add dynamic prior on top of RL and refinement (ablation) | Graph-GRPO · PMO (23 tasks) | AUC-top10: 17.950 -> 18.987 | Improved |
| add prescreening on top of RL, refinement and dynamic prior (ablation) | Graph-GRPO · PMO (23 tasks) | AUC-top10: 18.987 -> 19.270 | Improved |
| add RL training to DeFoG on Tree | DeFoG · Tree | V.U.N. (%): 73.5 -> 97.5 | Improved |
| add RL training to DeFoG on Planar | DeFoG · Planar | V.U.N. (%): 95.0 -> 95.0 | No clear effect |
| add RL training to DeFoG on Planar | DeFoG · Planar | ratio to train set: 3.2 -> 1.5 | Improved |
| use refinement instead of de novo generation with RL (ablation) | Graph-GRPO · Valsartan SMARTS | average top-10 score | Improved |
| use refinement instead of de novo generation with RL (ablation) | Graph-GRPO · Scaffold Hopping | average top-10 score | No clear effect |
| use Graph-GRPO instead of GDPO on braf docking score | Graph-GRPO · Protein docking braf | top-5% docking score: -11.197 -> -11.141 | Hurt |
| use Graph-GRPO instead of GDPO on parp1 hit ratio | Graph-GRPO · Protein docking parp1 | hit ratio (%): 9.814 -> 60.763 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeFoG — DeFoG: Discrete Flow Matching for Graph Generation (ICML 2025) · better on 3 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Flow-GRPO: Training Flow Matching Models via Online RL (NeurIPS 2025 · similarity 0.88)
- Graph Diffusion Policy Optimization (NeurIPS 2024 · similarity 0.88)
- DeFoG: Discrete Flow Matching for Graph Generation (ICML 2025 · similarity 0.87)
- Learning Mean Field Games on Sparse Graphs: A Hybrid Graphex Approach (ICLR 2024 · similarity 0.86)
- Group-Graph Policy Optimization for Long-Horizon Agentic Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
- DACA-GRPO: Denoising-Aware Credit Assignment for Reinforcement Learning in Diffusion Language Models (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
